OpenAI Safety Systems leader David Robinson resigns amid the week’s upheaval

David Robinson, a leader on OpenAI's Safety Systems team, has resigned from the company, Business Insider reports. An OpenAI spokesperson told BI on Friday that Robinson left last week, describing another moment of upheaval in the company's safety ranks.

Robinson had previously led the company's policy planning. His recent work sat on safety transparency, including helping develop and share system cards — the public write-ups labs publish about what a model can do and how it was tested. He did not immediately respond to requests for comment.

The timing is the point. Days earlier, OpenAI said it had "parted ways" with three researchers over sensitive-information policies, a personnel move we covered in full in OpenAI parts ways with three safety researchers. The same week, the company is still digging through rogue-agent incidents after Hugging Face and has already privately warned more than 100 outside organizations — the disclosure spine in OpenAI alerts 100 organizations about rogue-agent activity. Johannes Heidecke, OpenAI's safety head, also left earlier this year.

Robinson's own early-September note on X is the plain-English version of the bench story. Things at OpenAI "are indeed changing significantly by the day, but I do not know whether we are changing fast enough," he wrote.

A policy firing and a Safety Systems resignation are not the same event. One is the company enforcing its information rules. The other is a named leader who helped explain the models to the outside world deciding he is done. OpenAI can treat both as routine churn. Readers do not have to.

Sam Altman told the company's developer conference on Tuesday that safety and alignment need to stay ahead of model capabilities. That line is easier to deliver than to staff. When the people closest to the transparency paperwork leave in the same week the private warning list clears a hundred organizations, the public scorecard and the internal bench start to look like different products.

The unread page is not whether OpenAI can hire another Safety Systems lead. It is whether the lab can keep asking for trust on alignment while the people who wrote the system cards keep walking out under the same spotlight.